LogBasicCoreImpl.mesa
Copyright © 1985 by Xerox Corporation. All rights reserved.
Implements stream log reading, log tail management, random log reading.
Last edited by
MBrown on January 30, 1984 5:50:24 pm PST
Hauser, May 2, 1985 9:49:19 am PDT
Carl Hauser, April 10, 1986 3:45:22 pm PST
DIRECTORY
AlpineEnvironment,
AlpineInternal,
AlpineInline,
FileMap,
FilePageMgr,
AlpineLog,
LogBasic,
LogInline,
LogBasicInternal,
LogRep,
PrincOpsUtils,
Process,
RuntimeError;
LogBasicCoreImpl:
MONITOR
IMPORTS
AlpineInline,
FileMap,
FilePageMgr,
AlpineLog,
LogInline,
PrincOpsUtils,
Process,
RuntimeError
EXPORTS
LogBasic,
LogBasicInternal =
BEGIN
PageCount: TYPE = AlpineEnvironment.PageCount;
PageNumber: TYPE = AlpineEnvironment.PageNumber;
PageRun: TYPE = AlpineEnvironment.PageRun;
FileID: TYPE = AlpineEnvironment.FileID;
VolumeID: TYPE = AlpineEnvironment.VolumeID;
WordNumber: TYPE = LogBasic.WordNumber;
WordCount: TYPE = LogBasic.WordCount;
RecordID: TYPE = AlpineLog.RecordID;
nullRecordID: RecordID = AlpineLog.nullRecordID;
Comparison: TYPE = AlpineLog.Comparison;
wordsPerPage: CARDINAL = AlpineEnvironment.wordsPerPage;
CallerProgrammingError: ERROR = CODE;
InternalProgrammingError: ERROR = CODE;
Header:
PROC [p:
LONG
POINTER]
RETURNS [
LONG
POINTER
TO LogRep.Header] =
INLINE {
RETURN [LOOPHOLE[p]] };
Static log file state
logFileHandle: FileMap.Handle;
logFileSize: PageCount;
logFileLength: WordCount;
logChunkSize: INT = 4;
PageRunStartingWith:
PROC [p: PageNumber]
RETURNS [pr: PageRun] =
INLINE {
RETURN [[firstPage: p, count: MIN[logFileSize-p, logChunkSize]]] };
PageFollowing:
PROC [pr: PageRun]
RETURNS [p: PageNumber] =
INLINE {
p ← pr.firstPage+pr.count;
RETURN [IF p = logFileSize THEN 0 ELSE p] };
IsMemberOfPageRun:
PROC [p: PageNumber, pr: PageRun]
RETURNS [
BOOL] =
INLINE {
RETURN [p IN [pr.firstPage .. pr.firstPage+pr.count)] };
EstablishLogFile:
PUBLIC
PROC [
logVolume: AlpineEnvironment.VolumeID, logFile: AlpineEnvironment.FileID] = {
LogBasic.EstablishLogFile
logFileHandle ← FileMap.Register[volumeID: logVolume, fileID: logFile];
logFileSize ← FilePageMgr.GetSize[logFileHandle];
logFileLength ← AlpineInline.WordsFromPages[logFileSize];
};
LogFileSize:
PUBLIC
PROC []
RETURNS [PageCount] = {
RETURN[logFileSize];
};
LogUsage:
PUBLIC ENTRY PROC []
RETURNS [PageCount] ~ {
nPages: INT;
RETURN[IF (nPages ← (PageFollowing[chunkTable[tail].chunk.pageRun] - oldestFilePage)) >= 0 THEN nPages + nBufferPages ELSE nPages + nBufferPages + logFileSize];
};
AlpineLog tail.
This data structure keeps status information about log chunks near the log tail.
Specifically, any chunk that may be dirty but which has not yet been forced out
must be present in the table. The table is managed as a queue which always has
at least one element in it. When any chunk completes a forceout, its slot in the
table is marked free. A slot is reused if it is free and corresponds to the earliest
log page in the table.
Types
ChunkTableSize: CARDINAL = 4;
ChunkTableIndex: TYPE = [0..ChunkTableSize);
ChunkTableRec:
TYPE =
RECORD [
chunk: FilePageMgr.VMPageSet ← FilePageMgr.nullVMPageSet,
Contains a PageRun (PageNumber, Count) and the LONG POINTER to the mapped pages.
recordID: AlpineLog.RecordID ← AlpineLog.nullRecordID,
ID of first word of first page of chunk.
forceInProgress:
BOOL ←
FALSE
Set TRUE before forking process to force final update to the chunk; set FALSE when
this process finishes. So if chunk is not the tail and NOT forceInProgress, it is eligible
for re-use.
];
NextChunkTableIndex:
PROC [i: ChunkTableIndex]
RETURNS [ChunkTableIndex] =
INLINE {
RETURN [IF i = LAST[ChunkTableIndex] THEN FIRST[ChunkTableIndex] ELSE SUCC[i]] };
State
Variables below are only updated by AdvanceChunk, except that the TRUE -> FALSE
transition of forceInProgress for a chunk is made by ForceChunk
(forked by AdvanceChunk), and oldestFilePage is updated by Release.
chunkTable: ARRAY ChunkTableIndex OF ChunkTableRec;
tail: ChunkTableIndex;
completion:
CONDITION;
Wait here for a forceInProgress = FALSE in chunkTable.
curVersion: LogRep.PageVersion;
Version bit for pages in tail chunk
nextTailChunkID: AlpineLog.RecordID;
RecordID of first word of first page of NEXT tail chunk. Used to bounds-check RecordIDs
presented to Force and Get. Redundant; can be computed from chunkTable[tail] .
oldestFilePage: PageNumber;
The least-recently written page of the log file that cannot be overwritten.
enforceBufferPages: BOOL;
nBufferPages:
INT = 2 * AlpineInternal.MaxCoordinators;
If enforceBufferPages, then do not allow writing within nBufferPages pages
of oldestFilePage
Operations on the log tail.
AdvanceChunk:
PUBLIC
ENTRY
PROC []
RETURNS [
version: LogRep.PageVersion, pagesInChunk: INT, firstPagePtr: LONG POINTER] = {
LogBasicInternal.AdvanceChunk
Caller must hold the log tail monitor; hence only one caller at a time.
newTail: ChunkTableIndex;
Fork a process to write the current tail chunk.
chunkTable[tail].forceInProgress ← TRUE;
Process.Detach[FORK ForceChunk[index: tail]];
Wait, if necessary, for room in the chunk table to allocate the next tail chunk.
DO
newTail ← NextChunkTableIndex[tail];
IF NOT chunkTable[newTail].forceInProgress THEN EXIT;
WAIT completion;
ENDLOOP;
IF chunkTable[newTail].chunk.pages #
NIL
THEN
FilePageMgr.ReleaseVMPageSet[
vMPageSet: chunkTable[newTail].chunk, releaseState: clean, keep: TRUE];
{
Determine the page number of the first page, and the number of pages, in
the next chunk.
nextPage: PageNumber ← PageFollowing[chunkTable[tail].chunk.pageRun];
IF nextPage = 0 THEN curVersion ← 1 - curVersion;
Get new chunk, update chunk table
tail ← newTail;
chunkTable[tail].recordID ← nextTailChunkID;
chunkTable[tail].chunk ← FilePageMgr.UseLogPages[
fileHandle: logFileHandle, pageRun: PageRunStartingWith[nextPage]];
{
-- hack to avoid completely filling up log file
biasedOldestFilePage: PageNumber ← oldestFilePage - nBufferPages;
IF biasedOldestFilePage < 0
THEN
biasedOldestFilePage ← biasedOldestFilePage + logFileSize;
IF (enforceBufferPages
AND
IsMemberOfPageRun[biasedOldestFilePage, chunkTable[tail].chunk.pageRun]) OR
IsMemberOfPageRun[oldestFilePage, chunkTable[tail].chunk.pageRun]
THEN ERROR AlpineLog.WriteFailed;
};
nextTailChunkID ← LogInline.AddC[r: nextTailChunkID,
words: AlpineInline.ShortWordsFromPages[chunkTable[tail].chunk.pageRun.count]];
};
RETURN [
version: curVersion,
pagesInChunk: chunkTable[tail].chunk.pageRun.count,
firstPagePtr: chunkTable[tail].chunk.pages];
};
ForceChunk:
PROC [index: ChunkTableIndex] = {
On entry, chunkTable[index].forceInProgress.
Force the chunk, and set forceInProgress ← false.
Note: do NOT hold monitor during force out.
BroadcastCompletion:
ENTRY
PROC [] =
INLINE {
chunkTable[index].forceInProgress ← FALSE;
BROADCAST completion;
};
FilePageMgr.ForceOutVMPageSet[chunkTable[index].chunk];
BroadcastCompletion[];
};
ForceTo:
PUBLIC
PROC [followingRecord: RecordID] = {
LogBasicInternal.ForceTo
If the word before followingRecord is in the current tail chunk, force that chunk
(using this process). Then wait for other chunks that precede followingRecord to
go out.
TailForceRequired:
ENTRY
PROC []
RETURNS [doForce: BOOL, chunk: FilePageMgr.VMPageSet] = {
followingRecord must not be later than page about to be written.
IF AlpineLog.Compare[followingRecord, nextTailChunkID] = greater
THEN
RETURN WITH ERROR CallerProgrammingError;
IF AlpineLog.Compare[chunkTable[tail].recordID, followingRecord] = less
THEN {
FilePageMgr.ShareVMPageSet[chunkTable[tail].chunk];
RETURN [TRUE, chunkTable[tail].chunk] }
ELSE RETURN [FALSE, FilePageMgr.nullVMPageSet]
};
AwaitForce:
ENTRY
PROC [] = {
Return when all chunks that precede followingRecord are out.
FOR currentChunk: ChunkTableIndex
IN ChunkTableIndex
DO
IF AlpineLog.Compare[chunkTable[currentChunk].recordID, followingRecord] = less
THEN
UNTIL
NOT chunkTable[currentChunk].forceInProgress
DO
WAIT completion;
ENDLOOP;
ENDLOOP;
};
doForce: BOOL; chunk: FilePageMgr.VMPageSet;
[doForce, chunk] ← TailForceRequired[];
IF doForce
THEN {
FilePageMgr.ForceOutVMPageSet[vMPageSet: chunk];
FilePageMgr.ReleaseVMPageSet[vMPageSet: chunk, releaseState: clean, keep: TRUE] };
AwaitForce[];
};
OpenCoreForPut:
PUBLIC
ENTRY
PROC [
nextPage: PageNumber, version: LogRep.PageVersion, nextRecord: RecordID]
RETURNS [pagesInChunk: INT, firstPagePtr: LONG POINTER] = {
LogBasicInternal.OpenCoreForPut
enforceBufferPages ← FALSE;
curVersion ← version;
tail ← FIRST[ChunkTableIndex];
chunkTable[tail].recordID ← nextRecord;
chunkTable[tail].chunk ← FilePageMgr.UseLogPages[
fileHandle: logFileHandle, pageRun: PageRunStartingWith[nextPage]];
nextTailChunkID ← LogInline.AddC[r: chunkTable[tail].recordID,
words: AlpineInline.ShortWordsFromPages[chunkTable[tail].chunk.pageRun.count]];
RETURN [
pagesInChunk: chunkTable[tail].chunk.pageRun.count,
firstPagePtr: chunkTable[tail].chunk.pages];
};
AssertNormalOperation:
PUBLIC
ENTRY
PROC [] = {
enforceBufferPages ← TRUE;
};
CloseCoreForPut:
PUBLIC
ENTRY
PROC [] = {
LogBasicInternal.CloseCoreForPut
FOR i: ChunkTableIndex
IN ChunkTableIndex
DO
WHILE chunkTable[i].forceInProgress
DO
WAIT completion;
ENDLOOP;
FilePageMgr.ForceOutVMPageSet[chunkTable[i].chunk];
FilePageMgr.ReleaseVMPageSet[
vMPageSet: chunkTable[i].chunk, releaseState: clean, keep: FALSE];
chunkTable[i] ← [];
ENDLOOP;
};
WordNumberFromRecordID:
PUBLIC
ENTRY
PROC [thisRecord: RecordID]
RETURNS [result: WordNumber] = {
LogBasic.WordNumberFromRecordID
offsetFromNextTailChunk: WordCount;
IF AlpineLog.Compare[thisRecord, nextTailChunkID] = greater THEN GOTO tooLarge;
offsetFromNextTailChunk ← LogInline.WordsFromSubtract[
larger: nextTailChunkID, smaller: thisRecord ! RuntimeError.BoundsFault => GOTO tooSmall];
IF offsetFromNextTailChunk >= logFileLength THEN GOTO tooSmall;
result ← AlpineInline.WordsFromPages[
pages: PageFollowing[chunkTable[tail].chunk.pageRun]] - offsetFromNextTailChunk;
RETURN [IF result < 0 THEN result + logFileLength ELSE result];
EXITS
tooLarge => ERROR CallerProgrammingError;
tooSmall => ERROR CallerProgrammingError;
};
VerifyGet:
ENTRY
PROC [page: PageCount, pagePtr:
LONG
POINTER] = {
Called from Get if it encounters a page marked invalid.
IF Header[pagePtr].valid THEN RETURN;
IF IsMemberOfPageRun[page, chunkTable[tail].chunk.pageRun] THEN RETURN;
ERROR;
};
Release:
PUBLIC
PROC [beforeRecord: RecordID] = {
LogBasic.Release
AssignToOldestFilePage:
ENTRY
PROC [newOldestFilePage: PageNumber] =
INLINE {
oldestFilePage ← newOldestFilePage };
AssignToOldestFilePage[AlpineInline.PagesFromWords[
WordNumberFromRecordID[beforeRecord]]];
};
A readonly page stream on the log.
This stream has a limited capability for backup: it can save one page position that
MUST be returned to later.
End of log is determined at the page level.
nChunksReadAhead: CARDINAL = 4;
PageStreamObject:
TYPE =
RECORD [
chunk: FilePageMgr.VMPageSet ← FilePageMgr.nullVMPageSet,
savePageRun: PageRun,
pageInPageRun, savePageInPageRun: CARDINAL,
expectedVersion, saveExpectedVersion: LogRep.PageVersion,
pagePtr: LONG POINTER,
posSaved: BOOL,
nextReadAheadPage: PageNumber ];
pageStream: PageStreamObject;
OpenPageStream:
PROC [p: PageNumber]
RETURNS [endOfLog:
BOOL] = {
The initial stream position is NOT saved.
pageStream.chunk ←
FilePageMgr.ReadLogPages[
fileHandle: logFileHandle, pageRun: PageRunStartingWith[p]];
pageStream.pagePtr ← pageStream.chunk.pages;
pageStream.pageInPageRun ← 0;
pageStream.nextReadAheadPage ← PageFollowing[pageStream.chunk.pageRun];
THROUGH [1..nChunksReadAhead] DO ReadAhead[] ENDLOOP;
IF NOT Header[pageStream.pagePtr].valid THEN RETURN [TRUE];
pageStream.expectedVersion ← Header[pageStream.pagePtr].version;
pageStream.posSaved ← FALSE;
RETURN [FALSE];
};
ClosePageStream:
PROC [] = {
FilePageMgr.ReleaseVMPageSet[
vMPageSet: pageStream.chunk, releaseState: clean, keep: FALSE];
pageStream.chunk ← FilePageMgr.nullVMPageSet;
};
ReadAhead:
PROC [] = {
FilePageMgr.ReadAheadLogPages[
logFileHandle, PageRunStartingWith[pageStream.nextReadAheadPage]];
pageStream.nextReadAheadPage ←
IF pageStream.nextReadAheadPage < logFileSize - logChunkSize
THEN
pageStream.nextReadAheadPage + logChunkSize
ELSE 0;
};
SavePageStreamPos:
PROC [] = {
pageStream.savePageRun ← pageStream.chunk.pageRun;
pageStream.savePageInPageRun ← pageStream.pageInPageRun;
pageStream.saveExpectedVersion ← pageStream.expectedVersion;
pageStream.posSaved ← TRUE;
};
CurrentPageNumber:
PROC []
RETURNS [PageNumber] =
INLINE {
RETURN [pageStream.chunk.pageRun.firstPage + pageStream.pageInPageRun];
};
ForgetSavedPageStreamPos:
PROC [] = {
It is only valid to forget when positioned at the saved position.
IF NOT pageStream.posSaved THEN RETURN;
IF pageStream.savePageRun # pageStream.chunk.pageRun
THEN
ERROR InternalProgrammingError;
pageStream.posSaved ← FALSE;
};
AdvancePageStream:
PROC []
RETURNS [endOfLog:
BOOL] = {
pageStream.pageInPageRun ← pageStream.pageInPageRun+1;
IF pageStream.pageInPageRun = pageStream.chunk.pageRun.count
THEN {
p: PageNumber;
FilePageMgr.ReleaseVMPageSet[
vMPageSet: pageStream.chunk, releaseState: clean, keep: pageStream.posSaved];
p ← PageFollowing[pageStream.chunk.pageRun];
IF p = 0 THEN pageStream.expectedVersion ← 1 - pageStream.expectedVersion;
pageStream.chunk ←
FilePageMgr.ReadLogPages[fileHandle: logFileHandle, pageRun: PageRunStartingWith[p]];
pageStream.pagePtr ← pageStream.chunk.pages;
pageStream.pageInPageRun ← 0;
IF NOT pageStream.posSaved THEN ReadAhead[];
}
ELSE
pageStream.pagePtr ← pageStream.pagePtr + wordsPerPage;
RETURN[ (
NOT Header[pageStream.pagePtr].valid)
OR
(pageStream.expectedVersion # Header[pageStream.pagePtr].version) ];
};
SetPageStreamPosFromSavedPos:
PROC [] = {
IF NOT pageStream.posSaved THEN ERROR InternalProgrammingError;
IF pageStream.chunk.pageRun # pageStream.savePageRun
THEN {
FilePageMgr.ReleaseVMPageSet[vMPageSet: pageStream.chunk, releaseState: clean, keep: TRUE];
pageStream.chunk ←
FilePageMgr.ReadLogPages[fileHandle: logFileHandle, pageRun: pageStream.savePageRun];
pageStream.expectedVersion ← pageStream.saveExpectedVersion;
};
pageStream.pageInPageRun ← pageStream.savePageInPageRun;
pageStream.pagePtr ← pageStream.chunk.pages +
AlpineInline.ShortWordsFromPages[pageStream.pageInPageRun];
};
A readonly log record stream.
AlpineLog reading during recovery is significantly different than log reading for carrying out intentions. In the case of recovery, we expect to encounter the log end while reading; in carrying out intentions we expect all log records to be perfectly formed. Also, log reading for recovery is sequential, while log reading for intentions is more random. In the case of recovery we need the ability to first "peek" and then read; for intentions, reading is enough.
RecordStreamObject:
TYPE =
RECORD [
id: RecordID
of current log record, whose first word is on the current page stream page.
];
recordStream: RecordStreamObject;
OpenRecordStreamFromWord:
PUBLIC
PROC [firstWord: WordNumber]
RETURNS [notStartOfRecord: BOOL, currentRecord: RecordID] = {
notStartOfRecord ← OpenRecordStream[firstWord: firstWord,
firstRecord: LogInline.RecordIDFromWordNumber[firstWord]];
RETURN [notStartOfRecord, recordStream.id];
};
OpenRecordStreamFromCheckpoint:
PUBLIC
PROC [
checkpointWord: WordNumber, checkpointRecord: RecordID, firstRecord: RecordID]
RETURNS [notStartOfRecord: BOOL, currentRecord: RecordID] = {
offsetFromCheckpoint, firstWord: WordCount;
IF AlpineLog.Compare[checkpointRecord, firstRecord] = less
THEN
ERROR CallerProgrammingError;
offsetFromCheckpoint ← LogInline.WordsFromSubtract[
larger: checkpointRecord, smaller: firstRecord ! RuntimeError.BoundsFault => GOTO tooSmall];
IF offsetFromCheckpoint >= logFileLength THEN GOTO tooSmall;
firstWord ← checkpointWord - offsetFromCheckpoint;
IF firstWord < 0 THEN firstWord ← firstWord + logFileLength;
notStartOfRecord ← OpenRecordStream[firstWord: firstWord, firstRecord: firstRecord];
RETURN [notStartOfRecord, firstRecord];
EXITS
tooSmall => ERROR CallerProgrammingError;
};
OpenRecordStream:
PROC [firstWord: WordNumber, firstRecord: RecordID]
RETURNS [notStartOfRecord: BOOL] = {
p: PageNumber;
w: CARDINAL [0..wordsPerPage);
[page: p, wordInPage: w] ← AlpineInline.DecomposeWords[firstWord];
IF LogInline.WordInPageFromRecordID[firstRecord] # w
THEN
ERROR CallerProgrammingError;
IF OpenPageStream[p].endOfLog THEN GOTO invalidPage;
SavePageStreamPos[];
recordStream ← [id: firstRecord];
RETURN [VerifyCurrentRecordID[].notStartOfRecord];
EXITS
invalidPage => ERROR InvalidPage;
};
InvalidPage: PUBLIC ERROR = CODE;
CurrentWordNumber:
PROC []
RETURNS [WordNumber] = {
RETURN [AlpineInline.WordsFromPages[CurrentPageNumber[]] +
LogInline.WordInPageFromRecordID[recordStream.id]];
};
CloseRecordStream:
PUBLIC
PROC [] = {
ClosePageStream[];
};
VerifyCurrentRecordID:
PROC []
RETURNS [notStartOfRecord:
BOOL] = {
wordInPage: CARDINAL ← LogInline.WordInPageFromRecordID[recordStream.id];
{ w:
CARDINAL ← 0;
UNTIL w >= wordInPage
DO
w ← w + Header[pageStream.pagePtr+w].nWords;
ENDLOOP;
IF w > wordInPage THEN RETURN [TRUE];
};
RETURN [Header[pageStream.pagePtr+wordInPage].isContinuation];
};
CheckCurrentRecord:
PUBLIC
PROC []
RETURNS [truncated:
BOOL] = {
wordInPage: CARDINAL ← LogInline.WordInPageFromRecordID[recordStream.id];
IF VerifyCurrentRecordID[].notStartOfRecord THEN ERROR InternalProgrammingError;
IF Header[pageStream.pagePtr+wordInPage].hasContinuation
THEN {
IF (Header[pageStream.pagePtr+wordInPage].nWords + wordInPage) # wordsPerPage
THEN
GOTO checkFailed;
DO
IF AdvancePageStream[].endOfLog THEN GOTO truncatedRecord;
IF
NOT Header[pageStream.pagePtr].isContinuation
THEN GOTO truncatedRecord;
IF NOT Header[pageStream.pagePtr].hasContinuation THEN EXIT;
IF Header[pageStream.pagePtr].nWords # wordsPerPage THEN GOTO checkFailed;
ENDLOOP;
IF Header[pageStream.pagePtr].nWords
NOT
IN [1 .. wordsPerPage]
THEN
GOTO checkFailed;
}
ELSE {
IF Header[pageStream.pagePtr+wordInPage].nWords
NOT IN [1 .. wordsPerPage-wordInPage] THEN GOTO checkFailed;
};
SetPageStreamPosFromSavedPos[];
RETURN [FALSE];
EXITS
checkFailed => { SetPageStreamPosFromSavedPos[]; ERROR InternalProgrammingError };
truncatedRecord => { SetPageStreamPosFromSavedPos[]; RETURN [TRUE] };
};
GetCurrentRecord:
PUBLIC
PROC [currentRecord: RecordID, to: AlpineLog.Block]
RETURNS [status: AlpineLog.ReadProcStatus, wordsRead: CARDINAL] = {
wordInPage: CARDINAL ← LogInline.WordInPageFromRecordID[recordStream.id];
wordsRead ← 0;
IF LogInline.Compare[currentRecord, recordStream.id] # equal
THEN
ERROR CallerProgrammingError;
DO
Read one log block from this page.
wordsThisBlock, wordsThisCopy: CARDINAL;
hasContinuation, isContinuation: BOOL;
Read the header for the log block.
[hasContinuation: hasContinuation, isContinuation: isContinuation, nWords: wordsThisBlock] ←
Header[pageStream.pagePtr+wordInPage]^;
IF wordsRead > 0 AND NOT isContinuation THEN GOTO truncatedRecord;
wordsThisBlock ← wordsThisBlock-SIZE[LogRep.Header];
wordInPage ← wordInPage+SIZE[LogRep.Header];
DO
Copy a run of words to "to".
wordsThisCopy ← MIN[to.length, wordsThisBlock];
IF to.base #
NIL
THEN {
PrincOpsUtils.LongCopy[
to: to.base, from: pageStream.pagePtr+wordInPage, nwords: wordsThisCopy];
to.base ← to.base+wordsThisCopy;
};
wordsRead ← wordsRead+wordsThisCopy;
Now what?
IF wordsThisCopy # to.length
THEN {
to.length > wordsThisCopy = wordsThisBlock; go read next log block.
to.length ← to.length-wordsThisCopy;
IF NOT hasContinuation THEN GOTO sourceExhausted;
IF AdvancePageStream[].endOfLog THEN GOTO truncatedRecord;
wordInPage ← 0;
GOTO nextBlock;
}
ELSE
IF to.rest #
NIL
THEN {
wordsThisCopy = to.length AND to.rest # NIL, so not done copying.
to ← to.rest^;
wordsThisBlock ← wordsThisBlock-wordsThisCopy;
wordInPage ← wordInPage + wordsThisCopy;
}
ELSE {
wordsThisCopy = to.length AND to.rest = NIL, so destination full; return.
SetPageStreamPosFromSavedPos[];
RETURN [
IF (NOT hasContinuation AND wordsThisBlock = wordsThisCopy) THEN normal
ELSE destinationFull, wordsRead]
}
REPEAT
nextBlock => NULL;
ENDLOOP;
ENDLOOP;
EXITS
sourceExhausted => {
SetPageStreamPosFromSavedPos[];
RETURN [sourceExhausted, wordsRead]
};
truncatedRecord => {
SetPageStreamPosFromSavedPos[];
ERROR CallerProgrammingError;
};
};
GetCurrentPageAndVersion:
PUBLIC
PROC []
RETURNS [PageNumber, LogRep.PageVersion] = {
RETURN[CurrentPageNumber[], pageStream.expectedVersion];
};
AdvanceRecordStream:
PUBLIC
PROC []
RETURNS [endOfLog, truncatedRecord: BOOL, currentRecord: RecordID] = {
It is ok to call Advance when recordStream is in a bad state, e.g. not pointing to the start
of a record. It should be pointing to the start of a block, however. Note that
wordInPage = 0 is sufficient.
wordsInRecord: CARDINAL ← 0;
endOfLog ← truncatedRecord ← FALSE;
{ -- block for EXITS
wordInPage: CARDINAL ← LogInline.WordInPageFromRecordID[recordStream.id];
ForgetSavedPageStreamPos[];
WHILE Header[pageStream.pagePtr+wordInPage].hasContinuation
DO
wordsInRecord ← wordsInRecord + Header[pageStream.pagePtr+wordInPage].nWords;
wordInPage ← 0;
IF AdvancePageStream[].endOfLog
THEN {
endOfLog ← truncatedRecord ← TRUE; GOTO return };
IF
NOT Header[pageStream.pagePtr].isContinuation
THEN {
truncatedRecord ← TRUE; GOTO return };
ENDLOOP;
{ wordsInLastBlock: CARDINAL ← Header[pageStream.pagePtr+wordInPage].nWords;
wordInPage ← wordInPage + wordsInLastBlock;
wordsInRecord ← wordsInRecord + wordsInLastBlock };
IF wordInPage < wordsPerPage
AND
Header[pageStream.pagePtr+wordInPage].nWords = 0 THEN {
wordsInRecord ← wordsInRecord + (wordsPerPage-wordInPage);
wordInPage ← wordsPerPage;
};
IF wordInPage = wordsPerPage
AND AdvancePageStream[].endOfLog
THEN {
endOfLog ← TRUE;
};
GOTO return
EXITS return => {
SavePageStreamPos[];
recordStream.id ← LogInline.AddC[recordStream.id, wordsInRecord];
RETURN[endOfLog, truncatedRecord, recordStream.id] };
}};
Recovering from loss of the restart record.
This code checks several invariants of the log that are not checked when the restart record is used. In particular, a log with more than one "end" will be discovered by this code, but not when the restart record is used. So it may be useful to run this code during debugging of the system.
nTransitionPages: INT = 10 * logChunkSize;
LocateFirstRecord:
PUBLIC
PROC []
RETURNS [firstRecord: WordNumber] = {
notStartOfRecord: BOOL;
endOfLog: BOOL ← FALSE;
[notStartOfRecord: notStartOfRecord] ← OpenRecordStreamFromWord[
AlpineInline.WordsFromPages[LocateStartingPage[]]];
IF notStartOfRecord THEN [endOfLog: endOfLog] ← AdvanceRecordStream[];
IF endOfLog THEN ERROR InternalProgrammingError;
firstRecord ← CurrentWordNumber[];
CloseRecordStream[];
RETURN [firstRecord];
};
LocateStartingPage:
PROC []
RETURNS [PageNumber] = {
Scans the log once.
SkipAhead:
PROC [] = {
A scan is in progress, but EndOfLog has been seen.
Move forward nTransitionPages pages.
nextPage: PageNumber = CurrentPageNumber[] + nTransitionPages;
ClosePageStream[];
IF OpenPageStream[nextPage].endOfLog THEN ERROR InternalProgrammingError;
};
startPage, beginPage: PageNumber;
{ -- Read to find a page in a "clean" area.
IF OpenPageStream[0].endOfLog THEN GOTO skip;
THROUGH [0 .. nTransitionPages)
DO
IF AdvancePageStream[].endOfLog THEN GOTO skip;
ENDLOOP;
EXITS
skip => SkipAhead[];
};
startPage ← CurrentPageNumber[];
Next EndOfLog really IS the end of the log!
DO IF AdvancePageStream[].endOfLog THEN EXIT ENDLOOP;
Skip past any junk that might follow the EndOfLog, then save the page number.
SkipAhead[];
beginPage ← CurrentPageNumber[];
Now read back to startPage. Should see no more EndOfLogs.
UNTIL CurrentPageNumber[] = startPage
DO
IF AdvancePageStream[].endOfLog THEN ERROR InternalProgrammingError;
ENDLOOP;
ClosePageStream[];
RETURN [beginPage];
};
Random log reading.
Get:
PUBLIC
PROC [thisRecord: AlpineLog.RecordID, to: AlpineLog.Block]
RETURNS [status: AlpineLog.ReadProcStatus, wordsRead: CARDINAL] = {
LogBasic.Get
Enable UNWIND => ReleaseChunk[]
chunk: FilePageMgr.VMPageSet;
pagePtr: LONG POINTER;
pagesLeftInChunk: CARDINAL;
GetChunk:
PROC [nextPage: PageNumber] = {
chunk ←
FilePageMgr.ReadLogPages[
fileHandle: logFileHandle, pageRun: PageRunStartingWith[nextPage]];
pagePtr ← chunk.pages;
pagesLeftInChunk ← chunk.pageRun.count - 1;
};
ReleaseChunk:
PROC [] =
INLINE {
FilePageMgr.ReleaseVMPageSet[vMPageSet: chunk, releaseState: clean, keep: FALSE];
};
AdvancePage:
PROC [] =
INLINE {
IF pagesLeftInChunk = 0
THEN {
nextPage: PageNumber ← PageFollowing[chunk.pageRun];
ReleaseChunk[];
IF nextPage = 0 THEN pageVersion ← 1-pageVersion;
GetChunk[nextPage];
}
ELSE {
pagePtr ← pagePtr + AlpineEnvironment.wordsPerPage;
pagesLeftInChunk ← pagesLeftInChunk - 1;
};
};
isFirstPage: BOOL ← TRUE;
pageVersion: LogRep.PageVersion ← 0;
wordInPage: CARDINAL;
Verify:
PROC [] = {
Ensure that wordInPage points to a LogRep.Header, assuming that page
is properly formatted.
w: CARDINAL ← 0;
UNTIL w = wordInPage
DO
w ← w + Header[pagePtr+w].nWords;
IF w > wordInPage THEN ERROR InternalProgrammingError;
ENDLOOP;
};
{
firstPage: PageNumber;
[page: firstPage, wordInPage: wordInPage] ← AlpineInline.DecomposeWords[
WordNumberFromRecordID[thisRecord]];
GetChunk[firstPage];
};
Verify[];
wordsRead ← 0;
DO
Read one log block from this page.
wordsThisBlock, wordsThisCopy: CARDINAL;
hasContinuation, isContinuation: BOOL;
version: LogRep.PageVersion;
Read valid bit from the header for first block on this page.
IF
NOT Header[pagePtr].valid
THEN
VerifyGet[
chunk.pageRun.firstPage+(chunk.pageRun.count-(1+pagesLeftInChunk)), pagePtr];
Read the header for this log block.
[version: version, hasContinuation: hasContinuation, isContinuation: isContinuation,
nWords: wordsThisBlock] ← Header[pagePtr+wordInPage]^;
wordsThisBlock ← wordsThisBlock-LogRep.Header.SIZE;
wordInPage ← wordInPage+LogRep.Header.SIZE;
IF isFirstPage
THEN {
IF isContinuation THEN ERROR InternalProgrammingError;
pageVersion ← version;
isFirstPage ← FALSE }
ELSE {
IF
NOT isContinuation
OR (version # pageVersion)
THEN
ERROR InternalProgrammingError
};
DO
Copy a run of words to "to".
wordsThisCopy ← MIN[to.length, wordsThisBlock];
IF to.base #
NIL
THEN {
PrincOpsUtils.LongCopy[
to: to.base, from: pagePtr+wordInPage, nwords: wordsThisCopy];
to.base ← to.base+wordsThisCopy;
};
wordsRead ← wordsRead+wordsThisCopy;
Now what?
IF wordsThisCopy # to.length
THEN {
to.length > wordsThisCopy = wordsThisBlock; go read next log block.
to.length ← to.length-wordsThisCopy;
IF
NOT hasContinuation
THEN {
ReleaseChunk[];
RETURN[sourceExhausted, wordsRead]
};
AdvancePage[];
wordInPage ← 0;
GOTO nextBlock;
}
ELSE
IF to.rest #
NIL
THEN {
wordsThisCopy = to.length AND to.rest # NIL, so not done copying.
to ← to.rest^;
wordsThisBlock ← wordsThisBlock-wordsThisCopy;
wordInPage ← wordInPage + wordsThisCopy;
}
ELSE {
wordsThisCopy = to.length AND to.rest = NIL, so destination full; return.
ReleaseChunk[];
RETURN [
IF (NOT hasContinuation AND wordsThisBlock = wordsThisCopy) THEN normal
ELSE destinationFull, wordsRead]
}
REPEAT
nextBlock => NULL;
ENDLOOP;
ENDLOOP;
};
END.
OPEN PROBLEMS
Need a way to get the log chunk size from FilePageMgr (see logChunkSize constant above).
CHANGE LOG
Created by MBrown on June 15, 1982 3:38 pm
Introduce a separate monitor for "core" log data structures.
Changed by MBrown on September 10, 1982 9:45 pm
Bug: attempted to release an uninitialized chunk, in AdvanceChunk.
Now runs simple log read/write test.
Changed by MBrown on September 25, 1982 3:27 pm
First "complete" implementation of LogBasic.
Changed by MBrown on November 16, 1982 10:10 pm
Revised Get to be more closely analogous to GetCurrentRecord.
Changed by MBrown on April 3, 1983 4:17 pm
Test "checkpointRecord <= firstRecord" changed to checkpointRecord < firstRecord",
since equality is now ok.
Changed by MBrown on June 27, 1983 9:56 am
Added AssertNormalOperation proc. Now, if enforceBufferPages, log write
fails if within nBufferPages pages of wrap-around (in AdvanceChunk).
Changed by MBrown on January 30, 1984 10:18:19 am PST
Cedar 5.0: use PrincOpsUtils.
Hauser, March 8, 1985 10:44:53 am PST
Nodified, added copyright.
Carl Hauser, May 2, 1985 9:48:22 am PDT
changes to: LocateFirstRecord: Initialized endOfLog.
Carl Hauser, July 12, 1985 9:56:24 am PDT
changes to: LogBasicCoreImpl nBufferPages depends on number of coordinators to insure better chance for successful warmstart following a filling-up of the log.
Carl Hauser, October 4, 1985 1:31:01 pm PDT
Change "Log" to "AlpineLog"